Background

RFX (Regulatory Factor binding to the X-box) transcription factors (TFs) share and are defined by a conserved, specialized winged-helix type DNA binding domain (DBD) [1]. RFX genes have been identified in all animals within the Unikont branch of eukaryotes, which excludes algae, plants and various protozoan branches [2]. Metazoan genomes encode one to several RFX genes. C. elegans possesses one, Drosophila has two [3, 4], mammals have eight and – due to genome duplication – fishes have nine RFX genes [2, 5,6,7,8,9,10]. Human RFX1–7 have previously been described [9], while RFX8 (ENSG00000196460, www.ensembl.org) has not been characterized.

In different organisms, RFX TFs have been shown to regulate genes involved in various and seemingly disparate cellular and developmental processes [7] like the cell cycle and DNA repair [11, 12], or aspects of cellular differentiation, like the functional maturation of cells of the immune response [13] and the development of cilia on the surface of polarized cells [14,15,16]. As a consequence of these roles in development, mutations in RFX genes can lead to severe disease states. Mutations in RFX5 cause autosomal recessive Bare Lymphocyte Syndrome (OMIM #209920), characterized by severe combined immunodeficiency due to failure in HLA expression. Mutations in RFX6 cause autosomal recessive Mitchell-Riley Syndrome, characterized by neonatal diabetes and malformations of the gut (OMIM #615710). Rfx mutant mice exhibit a plethora of mild to fatal phenotypes, ranging from male sterility [17] to brain abnormalities [18]. These phenotypes are often attributed to cilia dysfunction [17, 19,20,21]. Of note, many human ciliopathy genes are strongly assumed to be RFX TF targets, given that their orthologs have been shown to be RFX TF targets in several different organisms, ranging from C. elegans to mouse [22,23,24].

In addition to the DBD, RFX TFs may contain other conserved domains like activation (AD) and dimerization (DIM) domains and the domains B and C of unknown function [7, 9]. The RFX DBD recognizes an imperfect inverted repeat sequence, the X-box motif, to which it binds [1]. RFX TF binding to the X-box motif has repeatedly been demonstrated by using methods ranging from in vitro binding studies, in vivo expression and mutation analyses to SELEX and ChIP sequencing approaches [8, 25,26,27,28]. Combined, these approaches led to the discovery of large batteries of RFX target genes [16].

By contrast, very little is known about upstream regulators of RFX genes. So far only a few studies in mice, zebrafish and flies have identified TFs of the bHLH class, Neurog3 and Atonal, as well as the homeobox protein Noto as upstream regulators of RFX genes [29,30,31]. In the yeast S. cerevisiae an upstream phosphorylation cascade controls expression of the RFX gene Crt1 [32].

In this study – using extensive analysis of data from the FANTOM5 database followed by experimental validations – we present an in-depth characterization of the entire human RFX gene family (RFX1–8), including the previously undescribed RFX8 and RFX transcript isoforms that encode TFs without DBD. We provide an updated grou** of human RFX TFs and show that RFX functional domain composition is independent of expression profile. Our exhaustive analysis of RFX expression in many different human tissues and cell types suggests that RFX tissue and cell type specificity arises mainly from differences in TSS architecture and not from different transcript isoforms. We determined with high precision the positioning of X-box motifs with respect to TSS locations of human RFX target genes. Using cluster analysis based on tissue and cell type specific expression profiles we link defined RFX family members with the target gene batteries they regulate. Further, we provide a first list of candidate upstream regulators of human RFX genes. The wealth of data we provide will serve as the basis for future studies of the role of RFX TFs in human development and disease.

Results

Expression of human RFX genes in different tissue types

Detailed expression profiles of the human RFX1–8 genes have not been described. We used data from the FANTOM5 database that is based on experimental expression profiling by CAGE technology across a wide spectrum of human biological samples. The expression level of a given CAGE TSS location is defined by an arbitrary unit, tags per million (TPM) [33]. We extracted 37 CAGE TSS locations for RFX1–8 from the FANTOM 5database and shortlisted these to 30 TSS locations by merging those which are in close proximity to each other and have similar expression profiles (cf. Methods). We then named these 30 TSS locations alphabetically, whereby promoter A (pA) is the highest expressed TSS. Expression of each RFX TSS is described in detail for human tissues, primary cells and cell lines (Additional file 1). The wealth of biological samples allows classifying the expression profiles for human RFX1–8 in different cell types.

A given RFX TSS location is considered as being expressed broadly if it is expressed at TPM > 5 in a large number and variety of tissues (n > 10). Conversely, a given RFX TSS location is considered as being expressed specifically if it is expressed at TPM > 5 in a small number of tissues of the same organ (n < 10). We found most TSS locations of RFX1–3, 5 and 7 to be expressed broadly in many tissue types whereas the TSS locations of RFX4 and RFX6 are all expressed in specific tissue types. pA@RFX4 is highly specific in brain and spinal cord tissues, while pB and pC@RFX4 are highly specific in testis. RFX6 TSS locations are all specifically expressed in the gastrointestinal tract (GI) (Additional file 2). We performed hierarchical clustering of the 30 RFX TSS locations based on their expression values (TPM) across 135 human tissue samples. Thereby we identified four major tissue clusters, namely immune system [34], gastrointestinal tract [35, 36], testis [37] and brain and spinal cord [18, 38,39,40,41], and two minor clusters, namely uterus and lung [42] (Additional file 2).

The expression of RFX8 is very low, making it the most elusive member of the human RFX family that has hitherto avoided detection. Here we identified RFX8 TSS locations with highest expressions in some tissues of the immune system (pA, pC, pD) and the gastrointestinal tract (pB, pE). However, the tissue expression values for pC, pD and pE were hard to distinguish from background noise (TPM < 1). In primary cells and cell lines, RFX8 TSS locations had higher expression values, with the most prominent expression in a Schwannoma cell line (Additional file 2).

Connecting TSS expression profiles to protein-coding transcript isoforms

FANTOM5 data allowed us to determine TSS locations and expression profiles. In order to connect the 30 RFX TSS locations described above to known transcript isoforms, we set a maximum distance limit of 50 nt between the TSS location and the nearest Ensembl protein-coding transcript with a complete open reading frame. We found that 18 of these 30 RFX TSS locations matched Ensembl protein-coding transcripts. The remainder (12) of the 30 RFX TSS locations were treated as novel transcript isoforms in human tissues. We selected seven of these as representatives for experimental validation by RT-PCR and sequencing (Table 1, Additional file 3: Table S1). The seven novel transcripts consist of (i) testis specific pC@RFX1 and pE@RFX3, (ii) broadly expressed and highest in brain pC@RFX3, pC@RFX5, pA@RFX7, pC@RFX7, and (iii) lowly expressed pA@RFX8.

Table 1 RFX1–8 expression data and novel transcripts
Full size table

Next, we assessed the full transcript sequences (from 5â€Č to 3’ UTRs) including their coding potential (from start to stop codons) from both the matched Ensembl protein-coding transcripts and the novel sequence-verified transcripts (Table S2 in Additional file 2). Representatives of the RFX1–8 transcripts are shown in Fig. 1a. We found that the majority of RFX transcript isoforms originating from the same gene encode identical proteins suggesting that tissue and cell type specificity arises mainly from differences in TSS architecture and regulation through their corresponding promoters. The exceptions are RFX1, RFX4 and RFX8 with isoforms encoding different protein variants. The testis specific pC@RFX1 transcript isoform encodes a shortened N-terminal region upstream of the activating domain (AD). RFX4 transcript isoforms have been extensively studied [43,44,45] and thus complement our results, where we found isoforms encoding different RFX4 protein variants. The RFX8 gene encodes a TF protein lacking a DNA binding domain (DBD) (ENSP00000401536, www.ensembl.org). Here, we experimentally validated RFX8 transcripts by sequencing cDNA from human brain total RNA and uncovered novel splicing patterns leading to alternative RFX8 protein variants, with and without DBD (Table S1 in Additional file 3).

Fig. 1
figure 1

Representative RFX transcripts grouped according to their functional domain compositions. a Representative RFX transcripts (to scale in nucleotides / nt) can be categorized based on the presence or absence of functional domains. Group 1 consists of RFX1, RFX2, and RFX3, which have all the domains. Group 2 consists of RFX4, RFX6 and RFX8, which have all domains but the AD. Group 3 consists of RFX5 and RFX7, which have only the DBD. Group 4 is novel, consisting of isoforms of RFX4 and RFX8, which lack the DBD. The start of the black bar marks the TSS position. Green and red arrows mark start and stop codon positions, respectively. The RFX protein domains encoded by these transcripts are AD (activation domain), DBD (DNA binding domain), B (domain B), C (domain C), and DIM (dimerization domain). They are indicated using color-coded boxes. The DBD (red box), which typically spans 222–225 nt (cf. Table S2 in Additional file 3) serves as a size marker. b, c RFX4 and RFX8 TSS locations illustrate best that RFX functional domain composition is independent of expression profile. They are connected to Ensembl protein-coding transcripts or shown as novel, validated transcripts (in red). Exon numbers refer to those in the corresponding Ensembl transcript IDs (distance and positions are not to scale). pA@RFX4 (red) belongs to the brain and spinal cord cluster, whereas pB and pC@RFX4 (green) belong to the testis cluster (cf. Additional file 2). The highest expressed tissues for pA and pB@RFX8 are thymus and medial frontal gyrus, respectively, and they are not color-coded because of their low expression levels in tags per million (TPM < 5) (Table 1)

Full size image

RFX functional domain composition is independent of expression profile

Human RFX TFs were previously categorized through phylogenetic analysis of their four functional domains outside the DBD: activating domain (AD), domain B, domain C and dimerization domain (DIM) [9, 10, 16]. Given the variation in coding potential of all 30 RFX1–8 transcript isoforms, we investigated whether there is a correlation between the presence or absence of certain RFX functional domains and the CAGE TSS expression profiles in human tissues. First, we categorized all RFX1–8 transcripts into four groups based on their functional domain structure: (i) Group 1: RFX1–3 with all known domains, (ii) Group 2: RFX4, RFX6 and RFX8 lacking the AD, (iii) Group 3: RFX5 and RFX7 with only the DBD, (iv) Group 4: RFX4 and RFX8 lacking the DBD (Fig. 1a). When we then compared these four groups to their respective TSS expression profiles, we did not find any indication that RFX TFs with similar domain composition would be expressed broadly or specifically in a certain tissue cluster, suggesting that the RFX functional domain composition is independent of expression profile. We analyzed RFX4 and RFX8 in more detail to illustrate this point (Fig. 1b, c).

Based on the FANTOM5 expression profiles, the gene RFX4 is highly tissue specific compared to other RFX genes. In our analysis, we connected three RFX4 TSS locations to three different Ensembl protein-coding transcripts (Fig. 1b). The longest, highest expressed transcript falls into Group 2 (lacking the AD); it is specifically expressed in the brain and spinal cord. The other two less expressed transcripts are both testis specific, whereby one belongs to Group 2 (lacking the AD) and the other one to Group 4 (lacking the DBD). The newly described gene RFX8 is the least expressed of all the RFX genes. We connected the two highest expressed RFX8 TSS locations with three possible transcripts (Fig. 1c). The same TSS can lead to different transcripts encoding protein variants, which either fall into Group 2 (lacking the AD) or Group 4 (lacking the DBD), suggesting an additional layer of gene regulation on top of the TSS architecture itself. Interestingly, RFX8 transcripts with DBD revealed that the RFX8 DBD is slightly shorter than the DBDs in RFX1–7. Whereby, multiple sequence alignments of RFX DBD amino acid sequences reveal that it is the least conserved N-terminal amino acids of the DBD that are missing in RFX8 (Figure S1 in Additional file 3).

Tissue and cell type specific clustering of RFX family members with the target genes they regulate

To correlate and eventually predict which RFX family member regulates which target gene in which human tissue and cell type, we compared and clustered the expression profiles of all RFX family members with direct RFX target genes. We selected from the literature a large number of validated direct RFX target genes in humans, as demonstrated either by a biochemical interaction between an RFX TF and the respective X-box promoter motif or by confirmation of the X-box function by mutation analysis (Table S3 in Additional file 3). We then extracted the CAGE TSS expression values (TPM) of these genes from the FANTOM5 database and performed unsupervised heat map clustering based on the correlations of expression values across 135 tissue types (Fig. 2).

Fig. 2
figure 2

Heat map of tissue expression clusters of RFX1–8 and their experimentally confirmed target genes in humans. Heat map of unsupervised hierarchical clustering of 30 TSS locations of RFX genes and 185 TSS locations of validated RFX target genes (with shorthand p for promoter) based on the expression values in tags per million (TPM) across 135 human tissue samples extracted from FANTOM5. The heat map color-code represent Pearson correlation values with a gradient of − 1 in dark blue/blue (negative correlation), 0 in white (zero correlation) and 1 in yellow/orange (positive correlation). The graph was generated by the heatmap.2::gplots [95] R package. RFX TSS locations tissue clusters (y-axis) are color-coded as described in Additional file 2. The tissue cluster divisions of RFX target genes (x-axis) are based on groups of tissues with the highest expression values (TPM) of the respective TSS locations. The term “other tissues” includes adipose, kidney, lung, seminal vesicle, skeletal muscle, throat and uterus

Full size image

We identified strong tissue specific RFX and target gene clusters, namely for testis (RFX1–4 and target genes GPR56, ALMS1, RFX1) and the gastrointestinal tract (RFX5–6 and target gene INS/insulin). We also observed strong differences between clusters of the immune system (RFX5) and the brain (RFX1, 3, 4, 7). This underscores that (i) the respective RFX family members regulate different sets of target genes as they are not co-expressed in a given tissue type, and (ii) for a given target gene RFX TFs can act as activators or as inhibitors. For brain tissue and cell types, RFX1, 3, 4 and 7 clustered tightly together, indicating a preference for these RFX family members to (co-) regulate the expression of brain-specific genes such as the ciliopathy/Alström syndrome gene ALMS1, the dyslexia candidate gene KIAA0319 or the gene MAP1A. Interestingly, another member of the brain cluster, pC@RFX2 (Table 1, Additional file 2), in the context of target genes clustered separately (Fig. 2), suggesting that in the brain RFX2 regulates a distinct set of target genes. Alternatively, RFX2 may interact with other RFX family members or other co-factors without preference as long as they are co-expressed in a given tissue and cell type.

X-box motif positioning in the human genome

RFX TFs regulate their target genes by binding to a conserved X-box motif in the promoter region. Previous X-box motif searches have typically been carried out using 1–3 kb sequence windows upstream of the TSS or ATG, such as in C. elegans [27], D. melanogaster [23], mouse and human [46]. To our knowledge, precise X-box motif positioning has not been characterized in the human genome. Thus, we determined the most likely positioning of functional X-box motifs in the promoter region, defined as 5000 bp upstream (− 5000) and 2000 bp downstream (+ 2000) in relation to TSS locations, of experimentally validated human RFX target genes.

To facilitate the search, we used two curated TF binding profiles for human RFX available in the JASPAR (2018) database [47]: RFX2 (MA0600.1) and RFX5 (MA0510.1) (Table S4 in Additional file 3). As a control, we selected a 10-fold larger random set of TSS locations across the human genome. Our search effort revealed that X-box hits are typically located very close to RFX target genes TSS locations (Fig. 3, Table S5 in Additional file 3). Based on search and find statistics the X-box positioning window can be further subdivided into a robust window of − 500 to + 500 bp and a permissive window of − 2300 to + 1400 bp. Using an independent search approach (the MEME suite FIMO software) [48] we confirmed these overall search and find parameters for human X-box motifs. Our analysis enhances the prediction power of future searches for functional X-box motifs, which relates to both upstream and downstream of TSS locations of candidate human RFX target genes and pinpoints their likely locations. Functional X-box motifs at larger distances from TSS locations (e.g at distal enhancers) are likely to be the exception rather than the norm (Table S6 in Additional file 3).

Fig. 3
figure 3

X-box motif position with respect to TSS locations. Density frequency of X-box motif positions with respect to the TSS locations of experimentally proven direct RFX target genes in humans (shown in blue) and a set of 10× random TSS locations from FANTOM5 (shown in red): TSS -5000 to + 2000 bp windows were scanned with two JASPAR RFX motifs (RFX2 MA0600.1 and RFX5 MA0510.1) with 80% threshold. We define a sequence window as “robust” by the area where the two curves with 95% C.I. smoothing do not overlap. We define a sequence window as “permissive” by the area where the two curves intersect

Full size image

Prediction of upstream RFX regulators using transcription factor binding site (TFBS) analysis

Identifying the upstream regulators of RFX genes will allow predicting the developmental and cellular niche that RFX TFs occupy. Thus, we searched for TF binding profiles over-represented in the promoter and enhancer regions of all 8 human RFX genes. We used search windows of − 5000 to + 2000 bp in relation to 30 RFX TSS locations and - 200 to + 200 bp around the midpoints of 13 significantly correlated candidate RFX enhancer sequences (extracted from Andersson et al. [49]; Table S7 in Additional file 3). We then scanned these regions with all the core vertebrate TF binding profiles present in the JASPAR 2016 database [47]. The enrichment for TF binding profiles was assessed against a 10-fold larger random set of human promoter and enhancer regions using the oPOSSUM3 tool [50].

We identified 19 over-represented TF binding profiles (Fig. 4) associated to the TFs SP2 (specificity protein 2) (JASPAR profile MA0516.1), E2F4 (E2 factor 4) (MA0470.1), KLF16 (Kruppel like factor 16) (MA0741.1), SP8 (specificity protein 8) (MA0747.1), SP3 (specificity protein 3) (MA0746.1), EGR3 (early growth response 3) (MA0732.1), ESR1 (estrogen receptor alpha) (MA0112.3), Creb5 (cAMP responsive element binding protein 5) (MA0840.1), ZNF740 (zinc finger protein 740) (MA0753.1), ATF7 (activating transcription factor 7) (MA0834.1), SOX21 (sex determining region Y-box 21) (MA0866.1), MZF1 (myeloid zinc finger 1) (MA0056.1 and MA0057.1), Tcfl5 (transcription factor like 5) (MA0632.1), KLF5 (Kruppel like factor 5) (MA0599.1), SP1 (specificity protein 1) (MA0079.3), EGR1 (early growth response 1) (MA0162.2), TFAP2C (transcription factor AP-2 gamma) (MA0815.1) and JDP2 (Jun dimerization protein 2) (MA0656.1). The full list of the TF binding profiles can be found in Additional file 4.

Fig. 4
figure 4

TF binding profiles in the promoter and enhancer regions of RFX genes. Distribution of all the z-scores of all the core vertebrate transcription factor binding site (TFBS) profiles in JASPAR 2016, with the search areas consisting of − 5000 to + 2000 bp with respect to the 30 RFX TSS locations and − 200 bp to + 200 bp from the mid-points of the RFX enhancers, against a background of a set of 10× random TSS locations and enhancers with identical window size and matching %GC distribution from FANTOM5. High-scoring or over-represented TF binding site profiles were computed as having z-scores above the mean + 2 x standard deviation (red dotted line)

Full size image

siRNA validation of RFX regulators

To test if any of the over-represented TF binding profiles can be linked to functional upstream regulation of RFX genes, we selected two TFs within the high-scoring TF binding profiles. We used siRNA knockdown of SP2 and ESR1 followed by qRT-PCR measuring the fold change of mRNA expression levels of RFX genes. We could successfully demonstrate knockdown of SP2 and ESR1 both at the mRNA level by qRT-PCR and at the protein level by immunoblotting (Fig. 5). The TF binding profiles of SP2 and ESR1 both scored clearly above the z-score threshold (Fig. 4). We selected the human MCF7 breast cancer cell line for which data are available in FANTOM5. In this cell line the genes SP2, ESR1, RFX1–3, − 5, and − 7 are expressed sufficiently high (TPM > 5), while the genes RFX4, − 6 and − 8 are not expressed (TPM = 0). We used efficiency-adjusted fold change quantification against scrambled (Scr) control siRNA normalized to the geometric mean of HPRT1 and HSPCB as two independent reference genes [51]. All the Ct levels of the test siRNA and Scr control siRNA can be found in Additional file 5.

Fig. 5
figure 5

siRNA validation of candidate RFX regulators. The genes SP2 and ESR1 represent the high-scoring group of candidate RFX regulators (cf. Fig. 4 and Additional file 4). In the MCF7 breast cancer cell line, amplification efficiency-adjusted mRNA fold change quantifications of RFX1, 2, 3, 5 and 7 were normalized to the geometric mean of HPRT1 and HSPCB, whereby a fold change equaling 1 describes an unchanged expression level. For this we used (a) SP2 siRNA versus scrambled (Scr) control siRNA knockdown, and (b) ESR1 siRNA versus scrambled (Scr) control siRNA knockdown. In (a, b) error bars represent SEM and fold-change statistical significance was calculated using the student two-sample t-test (***p-value ≀0.01, **p-value ≀0.05, *p-value ≀0.1). a, c SP2 siRNA knockdown was confirmed at both the mRNA and protein level and a significant up-regulation of RFX7 and down-regulation of RFX5 were observed. b, c ESR1 siRNA knockdown was confirmed at both the mRNA and protein level and significant up-regulations of RFX2, 3, 5 and 7 were observed. c Immunoblotting band intensities were quantified using ImageJ and normalized with the indicated loading controls

Full size image

We observed that siRNA knockdown of SP2 and ESR1 in human MCF7 cells resulted in a significant fold change in the mRNA expression levels of at least one of the RFX genes (Fig. 5). siRNA knockdown of SP2 resulted in both activating and inhibiting effects on RFX genes, whereby only RFX7 showed significant up-regulation. In contrast, siRNA knockdown of ESR1 revealed consistent inhibitory effects on all the RFX genes analyzed, with RFX2, − 3, − 5, and − 7 being significantly up-regulated. These data show that computational TFBS analyses of the promoter regions of RFX1–8 correctly identified functional upstream regulators of these RFX genes. Depending on the individual RFX gene these upstream regulators act either as activators or as repressors.

Discussion

We have exhaustively analyzed all eight members of the human RFX TF gene family (RFX1–8). By extracting and computationally analyzing large-scale experimental data sets, we were able to describe in detail RFX gene expression as well as the RFX gene regulatory landscape in many different human tissues and cell types, including in parts the experimental validation thereof. We provide the first detailed experimental characterization of RFX8 and of RFX isoforms without DBD. Further, we provide insight into upstream regulators of human RFX genes and determined the sequence windows in which – in most cases – human RFX TFs act as direct regulators of their target genes. Thereby we provide an in-depth catalogue and key resource for future work on the roles that RFX TFs play in human development and disease.

Our extensive survey of all the human RFX (1–8) gene expression profiles enabled us to carefully analyze all the transcript isoforms and determine the potential protein variants encoded by these isoforms. We ordered all expression profiles from low to high, from broad to tissue specific, made tissue and cell type assignments, including isoform correlations and non-correlations, and thereby were able to cluster the expression profiles for all the isoforms of all the human RFX genes. We found and experimentally validated that – typically – RFX gene TSS locations (of the same gene) would lead to the same protein variant, suggesting that it is mostly promoter and TSS architecture that gives rise to diversity in gene expression profiles. These results highlight the importance of studying non-coding regulatory regions of key genes involved in developmental processes such as cell type specification and differentiation. Exceptions include TSS locations for the gene RFX4 that were spread across a large genomic distance leading to transcript isoforms encoding different tissue specific protein variants.

Our work lead to an updated grou** of human RFX TFs and showed that RFX functional domain composition is independent of expression profile. We identified two RFX genes, RFX4 and RFX8, which can encode protein variants without DBD. The function of RFX protein variants without DBD is unclear. Possibly, they act as tissue-specific co-repressors, similar to SHP proteins [52]. This potential role is clearly inferred for RFX4, where such competitive co-repression may occur in the testis but not in the nervous system [44]. Transcript validation for the newly described gene RFX8 revealed the possibility for encoding protein variants with and without a DBD. For the protein variant with DBD, this domain would be slightly shorter as it is missing the least conserved N-terminal 20 amino acids. In addition to the overall low expression level of RFX8, it raises the question of RFX8 functionality. RFX8 was most prominently expressed in Schwannoma cells, suggesting a role for RFX8 in Schwann cell proliferation.

Given the central role that RFX TFs play during development (e.g. in the differentiation of cilia), we were interested in finding candidate upstream regulators of RFX genes. We used computational predictions based on over-represented TF binding profiles to find candidate upstream regulators of RFX genes and thereby infer the developmental pathways that RFX1–8 are part of. The over-represented TF binding profiles that our analysis uncovered are associated with TFs involved in (i) neural development (SP2, ESR1, Creb5, SOX21) [http://genome-euro.ucsc.edu/). In the case of the RFX8 DBD transcript validation, at least 100 white colonies were screened with PCR M13 vector primers prior to sequencing, given the overall low expression of the RFX8 gene. Sequences of primers and verified transcripts are listed in Tables S8 and S9, respectively, in Additional file 3.

Determination of RFX protein domains

Peptide sequences of human RFX1–3 protein domains (AD, DBD, B, C and DIM) as described previously [9, 10] were used to determine the corresponding domains in human RFX4–8 using the T-coffee protein sequence alignment program [91] (http://www.tcoffee.org/). Visualization of the RFX transcripts in Fig. 1a with the protein domain composition was done using IBS software [Human reference sequence

The human reference sequence used is the Human Feb. 2009 (GRCh37/hg19) Assembly.